ARTICLE IN PRESS

JOURNAL OF SOUND AND VIBRATION

Journal of Sound and Vibration 311 (2008) 13911408 www.elsevier.com/locate/jsvi

Active vibration suppression by pole-zero placement using measured receptances

John E. Mottersheada,, Maryam Ghandchi Tehrania, Simon Jamesa, Yitshak M. Ramb
a

Department of Engineering, University of Liverpool, Brownlow Hill, Liverpool L69 3GH, UK b Department of Mechanical Engineering, Louisiana State University, Baton Rouge, USA

Received 28 November 2006; received in revised form 15 October 2007; accepted 16 October 2007 Available online 26 November 2007

Abstract The paper addresses the problem of pole-zero assignment using the receptance method in active vibration control and has applications particularly in vibration absorption and detuning of structures to avoid resonance. An output feedback approach is described that makes use of measured receptances, there being no requirement at all for the M, C, K matrices usually obtained by nite elements. Therefore, in the controller design, the approximations, assumptions and other modelling errors are largely eliminated. In addition, the method does not require the use of model reduction techniques or the estimation of unmeasured states by an observer. An advantage of the output feedback method, over state feedback, is that collocated actuatorsensor arrangements become possible. However this is achieved at the expense of creating characteristic equations nonlinear in the control gains. Numerical examples are provided to illustrate the working of the method. This is followed by a series of experimental tests carried out using collocated accelerometers and inertial actuators on a T-shaped plate. In a series of experiments poles and zeros are assigned separately and simultaneously. Stability robustness is demonstrated by applying a constraint to the singular values of the matrix return difference. r 2007 Elsevier Ltd. All rights reserved.

1. Introduction The problem of eigenvalue assignment was studied quite extensively within the active control community during the 1970s and 1980s. Wonham [1] had shown in 1967 that if a system was controllable, then its eigenvalues (or poles) could be assigned by appropriate choice of state feedback. Davison [2] determined the conditions under which output feedback could be applied for eigenvalue assignment. Kimura [3] studied the problem of incomplete state observation and Moore [4] considered the freedom offered by state feedback beyond the specication of distinct closed-loop eigenvalues. Kautsky et al. [5] described numerical methods for determining robust (well conditioned) solutions to the state-feedback pole-assignment problem by dening a solution space of linearly independent eigenvectors, corresponding to the required eigenvalues. The solutions obtained were such that the sensitivity of assigned poles to perturbations in the system and gain
Corresponding author.

E-mail address: j.e.mottershead@liverpool.ac.uk (J.E. Mottershead). 0022-460X/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.jsv.2007.10.024

ARTICLE IN PRESS
1392 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

matrices was minimised. More recently, Datta et al. [6] developed the closed-form solution for the partial pole-assignment problem where some desired eigenvalues were relocated while keeping all the other eigenvalues unchanged. In vibration analysis the purpose of assigning poles and zeros is to suppress vibration. Examples include, moving poles further to the left-hand side of the complex plane to increase damping; assigning a zero at the tuned frequency of a classical vibration absorber; and moving natural frequencies away from resonance with applied harmonic loads by applying a passive structural modication. There are in fact numerous examples of structures having natural frequencies close to xed excitation frequencies. Recently much attention has been focussed on inverse eigenvalue problems for assigning natural frequencies and mode shapes [7], vibration nodes [8] and anti-resonances [9] by structural modication. For example Mottershead et al. [10,11] measured rotational receptances and applied them to assign the natural frequencies and anti-resonances of a G-shaped structure by means of an added beam. Mottershead and Tehrani [12] carried out the structural modication of a helicopter tailcone also using measured rotational receptances obtained by means of an X-block attachment. Mottershead and Ram [13] reviewed the eld of inverse eigenvalue problems for vibration absorption by structural modication and active control. The advantage of the structural modication approach is that the system is guaranteed to remain stable. However there are very considerable disadvantages: (1) the form of the modication that can be realised physically (symmetry, positive-deniteness, pattern of non-zero matrix terms) is restrictive, (2) rotational receptances are very difcult to measure and require high levels of specialist expertise, and (3) the number of eigenvalues to be assigned must be matched by the rank of the modication. In recent times many of the disadvantages that prevented the practical application of active control to exible structures, with low damping and large numbers of modes, have been overcome. Inertial actuators for the application of sky-hook damping [14] and piezo-electric devices [15] for thin plate-like applications have been developed. Active damping by velocity feedback has received much attention recently. A good example is the smart panel described by Gardonio et al. in a three-part paper [1618] covering the theory, design and application of the system, which comprised 16 decentralised units for the control of sound transmission. Each control unit consisted of a collocated accelerometer-sensor and a piezoceramic patch actuator with a single channel velocity feedback controller to generate active damping. A different approach, active constrained layer damping [19], can be particularly helpful in damping the higher-frequency closed-loop modes that might otherwise become unstable due to spillover [20]. The state-of-art in structural vibration control however is the independent modal-space control method developed by Meirovitch and his students and described in the book Dynamics and Control of Structures [21]. This approach allows in principle for the control of one structural mode independently of the others. In practice, the control force must be applied in the physical coordinates, which means that sufcient actuators must be used to ensure that the selected mode is controllable while the others are unaffected by the control force. Modern distributed actuators may be designed so that the vibration modes of a plate are controlled selectively in this way [22], but these techniques are not applicable to large-scale built-up structures consisting of many components. Meirovitchs analysis is presented in terms of the physical mass, damping and stiffness, M, C, K, matrices nally arranged in the form of rst-order state-space equations (when general viscous damping is included). From the point of view of the structural dynamics it is preferable to work with the second-order matrix pencil [23]. Redening the second-order equations of motion into a rst-order realisation destroys the desirable physical matrix properties of symmetry, deniteness and bandedness, and consequently the rst-order state-space model does not preserve any notion of the second-order nature of the system. In most cases the M, C, K matrices are obtained by nite elements, and may include the representation of distributed piezo-electric actuators and sensors, as described for example by Lim et al. [24]. Another way of implementing the independent modal-space control of Meirovitch is to use modal test data, derived from measured receptances, as described by Stobener and Gaul [25] in the active vibration control of a car body. Rayleigh (proportional) damping was assumed. An alternative approach that uses the measured receptances directly and makes no assumption about damping was proposed by Ram and Mottershead [26]. They showed, in principle, how state-feedback control using measured receptances from the original (open-loop) system could be used to assign all the poles of the closed-loop system using just a single actuator. Additionally, this method preserved the second-order nature of the physical system since the receptances are given by, H(io) (o2M+ioC+K)1 and the poles and zeros (anti-resonances) were assigned without the need to

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1393

know or evaluate the M, C, K matrices. A further signicant point is that the familiar state-space approach requires a dynamic stiffness model and all the states must be measured, or estimated using an observer, in order to make the system equations complete. On the other hand, when using the receptance method only the available states are needed to complete the equations. This means that by the receptance method there is no need to estimate the unmeasured states using an observer. Neither is there any need for model reduction. In this paper the receptance method for output-feedback is described in detail. The very considerable advantage of the output feedback method over state feedback is that it allows the use of collocated actuators and sensors in multiple-inputmultiple-output (MIMO) systems. It is well known that the transfer function of a lightly damped system with a collocated actuatorsensor pair displays a line of interlacing poles and zeros just to the left of the imaginary axis. The benets of this arrangement were discussed in detail by Preumont [27] who showed the trajectory of one such pole to be into the left-hand half-plane, therefore remaining stable under closed-loop control. In addition to velocity feedback, for active damping, the method uses displacement feedback for active stiffness, thereby enabling the assignment of both poles and zeros to desired locations in the complex s-plane. The assignment of zeros is of special interest in vibration analysis because the vibration can be made to vanish at chosen frequencies and locations. Active stiffness is more difcult to achieve than active damping [28] as can be appreciated when considering the location of a modal circle on the Nyquist diagram. In the case of velocity feedback the modal circle is conned to the right-hand half of the diagram, whereas for displacement feedback the circle occupies the lower two quadrants and can therefore be very close to the threshold of instability at 1. The method, described in the following section, is applied to both numerical and experimental examples. Three experiments on a T-shaped plate are described. In the rst experiment, four poles are assigned, and in the second, two zeros are assigned, each to the two-pointreceptances corresponding to the locations of collocated proof-mass actuators and accelerometers. In the third experiment poles and zeros are assigned simultaneously. In all the experiments, the other poles of the system at higher frequencies are found to remain stable. 2. Poles and zeros The dynamical equations, in the Laplace frequency domain, may be cast in the form of the second-order matrix equation, s2 M sC Kxs Bus ps,
nn T T T

(1)

; M M ; C C ; K K ; M40; C, KX0 are the usual structural stiffness, damping where M, C, KAR and mass matrices, BARn m is the control force distribution matrix, x(s), p(s)ARn 1 represent the displacement states and external forces, respectively, and u(s)ARm 1 is the control force. Likewise, the output equation may be written as ys Dxs, where DAR expressed as
ln

(2)
l1

is the sensor distribution matrix and the output is denoted by yAR us G sFys,

. The feedback law is (3)

so that the output and rate gains are given by the terms in the matrices G,FARm l. Then, combining Eqs. (1)(3) leads to, s2 M sC K BG sFDxs ps. Our purpose is to assign poles lj of the closed-loop system determined by the solution of,   det l2 M l C BFD K BGD 0; j 1; 2; . . . ; r ; r p 2n j j (4)

(5)

which we assume to be distinct and closed under conjugation. A necessary and sufcient condition is that the matrices M, (C+BFD) and (K+BGD) should be real.

ARTICLE IN PRESS
1394 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

The closed-loop zeros, mk, of the ppth point receptance, denoted hpp(s), are given by the solution of a different eigenvalue problem,   det m2 (6) k Mp mk C BFDp K BGDp 0, where Mp is the submatrix corresponding to the ppth term of M. Thus, " # h mpp mT p mp1 . . . mp;p1 mp;p1 ; mT M p mp Mp and m11 6 . 6 . 6 . 6 6 mp1;1 6 Mp 6 6 mp1;1 6 6 . 6 . 4 . mn1 2 m1;p1 . . . mp1;p1 mp1;p1 . . . mn;p1 m1;p1 . . . mp1;p1 mp1;p1 . . . mn;p1 i

...

mpn

(7,8)

3 m 1n 7 . . . 7 7 7 mp1;n 7 7 7 mp1;n 7 7 7 . . . 7 5 mnn

(9)

with similar denitions for the submatrices (C+BFD)p and (K+BGD)p. It is seen that the zeros are the eigenvalues of the system grounded at the pth coordinate. As with the poles, we assume the zeros to be distinct and closed under conjugation. 3. Pole and zero assignment by using receptances The receptance matrix of the open-loop system is now dened as Hs s2 M sC K1 . Premultiplying both sides of Eq. (4) by H(s) then gives, I HsDZsxs Hsps; where DZs BG sFD xs I HsBG sFD1 Hsps or xs adjI HsBG sFD Hsps: detI HsBG sFD (14) (12) and nally the closed-loop receptance equation can be expressed in terms of the open-loop receptances as (13) (11) (10)

The closed-loop system poles may be assigned by selecting real-valued gains G, F, to satisfy the nonlinear characteristic equations, det I Hlj BG lj FD 0; j 1; 2; . . . ; r; rp2n, (15) where the assigned poles are distinct and closed under conjugation. Since the equations are nonlinear in the gains there may be one or more strictly real solutions G, F or there may be no solution, in which case the closed-loop system is uncontrollable. The zeros of the closed-loop system occur when terms in the numerator matrix product, adj(I+H(s)B(G+sF)D)H(s), vanish to zero. The zeros of the ppth receptance may therefore be assigned by selecting gains such that, adj I Hmk BG mk FD Hmk pp 0. (16)

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1395

We consider zeros that are distinct and closed under conjugation, so that if a solution (or solutions) exist then the gains are found to be strictly real. If the sensors and actuators are collocated, D BT 2 <mn and the gain matrices are symmetric and positive semi-denite, G GT 2 <mm ; F FT 2 <mm ; G; FX0, (18) (17)

then it is seen from Eq. (5) that the closed-loop system matrices are symmetric with unchanged deniteness properties, which means that the poles have strictly negative real parts. In practice the actuator and sensor dynamics, neglected in the above theory, must be considered, as indeed they are in the experimental study that follows in Section 5. The use of collocated actuatorsensor systems was discussed in detail by Preumont [27]. The resulting characteristic equations (Eqs. (5) and (6) combined with Eqs. (17) and (18)) retain considerably greater freedom in the choice of control gains than does a passive structural modication by the adjustment of physical parameters such as beam cross sections or added masses. In the following numerical and experimental examples the gain matrices are diagonal, F diag f i ; G diaggi ; i 1; 2; . . . ; m. (19) This is perhaps the simplest form for the gain matrices, but it should be noted that in more complex or larger structural problems there may be advantages in retaining the fully populated symmetric form of Eqs. (18). In principle, the fully populated matrices should allow more poles and zeros to be assigned than the number of actuators and sensors. Alternatively, a greater number of solutions to the nonlinear characteristic equations may become available, thereby allowing the selection of gains that result in the least cost of control or that render those assigned poles and zeros least sensitive to small changes in the gain terms. 4. Numerical examples The method is illustrated by a series of eigenvalue assignment exercises based upon the system shown in Fig. 1. The parameters of the system have the following values: k1 3, k2 2, k3 2, k4 1, c1 0.1, c2 0.1, m1 2, m2 1, m3 3. Example 1 (Assignment of poles). Two pairs of complex conjugate poles at l1,2 0.0170.7i and l3,4 0.0671.8i are assigned using two actuators supplying feedback control forces at masses m1 and m3. The mass, damping and stiffness matrices are given by 2 3 2 3 0 :1 0 0 2 0 0 6 7 6 7 0 :1 0 :1 5 ; M 4 0 1 0 5; C 4 0 0 0 3 0 0:1 0 :1 6 6 K4 2 1 2 2 4 2 3 1 7 2 5 3

f1 k1 m1 c1 x1 k2 m2 x2 c2 k4 k3 m3

f3

x3

Fig. 1. Three degree-of-freedom system.

ARTICLE IN PRESS
1396 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

and the matrices B and D are chosen to be, DB

T

1 0

0 0

0 1

! ,

so that y1 x1, y2 x3. The open-loop system receptances are at the values of s corresponding to the chosen eigenvalues,  1 H lj M l2 C l K ; j 1; . . . ; 4. j j Four characteristic equations (17), generally nonlinear in the gains, gi, fi, are then solved numerically using a GaussNewton method, det I Hlj B diag f i lj gi BT 0; j 1; . . . ; 4 with the result that, F diag0:0999; 0:0475; G diag2:3873; 0:3401:

In order to validate the results, the eigenvalues of the closed-loop system are obtained by the state-space method, " # 0 I A , M1 K B diaggi BT M1 C B diagf i BT where 2 3 0 7 0 :1 5 0:1475 3 1 7 2 5 3:3401

0:1999 6 T C B diagf i B 4 0 0 and 2

0 0 :1 0:1

8:3873 6 T K B diaggi B 4 2 1 which yields the following poles:

2 4 2

l1;2 0:0100 0:7000i; l3;4 0:0600 1:8000i; l5;6 0:0546 2:3599i: The rst two pairs exactly replicate the values assigned and the third pair of poles are stable. The initial and modied receptances at m1 are plotted in Fig. 2, represented by the solid (blue) and dashed (red) line, respectively. Example 2 (Assignment of zeros). The zero assignment is considered for the point receptance h22. Two pairs of complex conjugate zeros m1,2 0.02571.2i and m3,4 0.03772i are assigned using the feedback control forces at masses m1 and m3. The sets of nonlinear equations (18) are solved to obtain the gains fi and gi adj I Hmj B diaggi mj f i BT Hmj 22 0; j 1; . . . ; 4 which yields F diag0:0493; 0:048; G diag1:8691; 1:5223:

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1397

The zeros, being the eigenvalues of the system grounded at the second coordinate, may be obtained by the state-space method using the gains determined above. Thus, " A 1 T M 2 K B diaggi B 2 0 , T 1 M 2 C B diagf i B 2 I #

102 Amplitude

100 102 104 0 0 50 1 2 3 4 5

Phase

100 150 200 0 1 2 3 4 5 Frequency (rad/s)

Fig. 2. Example 1initial receptance (solid line) and modied receptance (dashed line).

102 Amplitude

100

102 0 0 50 Phase 100 150 200 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Frequency (rad/s)
Fig. 3. Example 2initial receptance (solid line) and modied receptance (dashed line).

0.5

1.5

2.5

3.5

4.5

ARTICLE IN PRESS
1398 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

where C B diagf i B
T

0:1493 0

0 0:1480

and K B diaggi BT
2

7:8691 1

1 4:5223

which yields the zeros, m1;2 0:025 1:2i; m3;4 0:037 2i; for the point receptance h22. The initial and modied receptances, plotted in Fig. 3, are represented by the solid (blue) and dashed (red) lines, respectively. Example 3 (Assignment of poles and zeros together). Poles and zeros are assigned to h11 at l1,2 0.0270.7i and m1,2 0.00870.9i, respectively. Four nonlinear equations are solved simultaneously det I Hl1 B diagf i l1 gi BT 0, det I Hl2 B diagf i l2 gi BT 0, adj I Hm1 B diaggi m1 f i BT Hm1 11 0, adj I Hm2 B diaggi m2 f i BT Hm2 11 0 which yields to the following gains: F diag0:4023; 0:0404; G diag0:3720; 0:6836:

102 Amplitude 100 102 104 0 0 50 Phase 100 150 200 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Frequency (rad/s)
Fig. 4. Example 3initial receptance (solid line) and modied receptance (dashed line).

0.5

1.5

2.5

3.5

4.5

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1399

State-space analysis, using these gains, results in the poles, l1;2 0:0200 0:7000i; l3;4 0:0992 1:6497i; l5;6 0:0798 2:2755i and zeros, m1;2 0:0080 0:9000i; m3;4 0:0654 2:1007i: The initial and modied receptances h11 are plotted in Fig. 4, represented by the solid (blue) and dashed (red) line, respectively.

250 mm 3 mm 60 mm 30 mm 12.5 mm 100 mm

50 mm
Fig. 5. T-shaped plate: (a) dimensions and (b) experimental arrangement.

Fig. 6. (a) First mode (stem bending) and (b) second mode (stem twisting).

ARTICLE IN PRESS
1400 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

5. Experimental examples Experiments were carried out on the T-shaped plate shown in Fig. 5 using two sets of collocated sensors (Kistler accelerometer type 8636C50) and inertial actuators (Micromega Dynamics type IA-01). The internal active damper that forms part of the inertial actuator, based on analogue integration of acceleration was not used. Instead, signals from the Kistler accelerometers were integrated twice by digital means, thereby enabling both velocity and displacement feedback, implemented using MATLAB/Simulink and dSPACE. Open-loop receptances were measured by means of a modal test using hammer excitation with the inertial actuators in place but not operational. The H1 estimator was applied using the following test parameters: sample rate 256 Hz, frequency resolution 0.125 Hz, number of hits 20 and an exponential window with a decay to 1% applied to the measured accelerations. The rst two natural frequencies of the open-loop system were at 40 Hz (252 rad/s) and 52 Hz (325 rad/s). The rst mode was a stem bending mode and the second showed stem twisting with the two arms in anti-phase. The third natural frequency, in-phase arm bending, occurred at 125 Hz (785 rad/s) some distance away from the rst two modes. The rst two mode shapes are shown in Fig. 6.
(a) 103

Amplitude

104

150

200

250 300 350 Frequency (rad/s)

400

450

(b) 103

Amplitude

104

150

200

250 300 350 Frequency (rad/s)

400

450

Fig. 7. Rational fraction curve t (a) h11(io), (b) h12(io) measurementsolid line; tted curvedashed line.

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1401

The purpose of using receptances instead of the M, C, K matrices is that uncertainties, approximations and assumptions associated with nite element models are avoided completely. However H(io) is available from vibration experiments and not H(s), as required by the theory. Rational fraction polynomials were tted to the measured terms in H(io) and the coefcients of the numerator and denominator polynomials determined [29]. The coefcients were found by solving a least-squares problem, which should be well conditioned so that the coefcients are not sensitive to small changes in the measurements. The receptances h11(io) and h22(io) were almost identical because of geometric symmetry of the T-plate and due to linearity h12(io) and h21(io) were found to be very similar. The tted receptances are presented in Fig. 7 where measurements are represented by full lines (blue) and tted curves are shown as dashed lines (red). The good agreement shown in
Force/Voltage (Amplitude) 0.5 0.4 0.3 0.2 0.1 0 0 Force/Voltage (Phase) 200 100 0 100 200 0 100 200 300 Frequency (rad/s)
Fig. 8. Actuator forcevoltage transfer function.

100

200

300

400

500

600

400

500

600

103 Receptance (Amplitude)

104

200

250

300 350 Frequency (rad/s)

400

Fig. 9. Assignment of poles. Solid line: open-loop receptance. Dashed line: closed-loop receptance for the rst test. Dot-dashed line: closed-loop receptance for the second test.

ARTICLE IN PRESS
1402 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

Fig. 7, for amplitude, was similarly obtained for phase. The following polynomials were identied: h11 s h22 s 7:956 1010 s2 1:382 108 s 6:438 105 , 1:476 1010 s4 4:987 109 s3 2:515 105 s2 0:0003862s 1 1:334 1010 s2 4:678 109 s 5:208 106 . 1:476 1010 s4 4:987 109 s3 2:515 105 s2 0:0003862s 1

h12 s h21 s

It should be pointed out that the rational fraction polynomials represent a model of the system. However all that is required of this model is that it is accurate at the chosen location of the poles and zeros to be assigned. In the case of lightly damped systems it is likely that the identied rational fraction polynomial will be accurate if the curve t agrees closely with the measured terms in H(io). An expression for the forcevoltage transfer function of an inertial actuator with a xed base is given by Preumont [27]. Above a critical frequency it is shown that the actuator behaves as an ideal force generator
(a) 103
Unmodified H11 Zero assignment at 10 + 300i

Amplitude 104

200

250

300 Frequency (rad/s)

350

400

(b) 103 Receptance (Amplitude)

Unmodified H22 Zero assignment at 10 + 300i

104

200

220

240

260

280 300 320 340 Frequency (rad/s)

360

380

400

Fig. 10. Assignment of zeros: (a) h11 and (b) h22.

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1403

103 Receptance (Amplitude)

Unmodified Zero Assignment at 15+ 295i , Pole Assignment at 12 + 265i

104

200

250

300 Frequency (rad/s)

350

400

Fig. 11. Simultaneous assignment of poles and zeros.

(a) 0

10 Imaginary

20

30

40

50 20

10

10 Real

20

30

40

(b) 0.2 0.1 0 0.1 0.2 0.5

0 Real

0.5

Fig. 12. Polar plot of det[I+0.37 H(io)B diag(gi+iofi)BT]: (a) full polar plot and (b) magnied view close to the origin.

ARTICLE IN PRESS
1404 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

with constant amplitude and zero phase. This is the case shown in Fig. 8, the force being measured by a force sensor inserted between the base of the actuator and a heavy rigid mass. Preumonts analysis is in fact a simplication of the real situation since the actuator was xed to the exible T-plate in our control application. However a small distance away from the T-plate resonances the expression was found to hold good. A gain of 0.37 was obtained from the curve shown in Fig. 8, being the average value of the almost constant pure gain (zero phase) at frequencies greater than the natural frequency of the actuator at around 65 rad/s. Experiment 1 (Assignment of poles). Poles were rstly assigned at l1,2 127284i and l3,4 227365i and then in a second test at l1,2 107290i and l3,4 257375i. In all of the three experiments the force and sensor distribution matrices were set to B DT I, where I denotes the identity matrix. In the rst test, gains with the values of G diag(205,10955) and F diag(4.4,11.8) were found and in the second test G diag(124,14339) and F diag(0.5,11.19). Fig. 9 shows experimental receptances for the open-loop system as the full line (blue) and closed-loop receptances, h11, for the rst and second tests. The receptance for the rst test is shown as a dashed line (red) and for the second test as a dot-dashed line (green). As expected, the peaks of the dashed and dot-dashed lines can be seen to agree very well with the imaginary parts of the assigned poles for the rst and second test, respectively. The actuators were operational during the closed-loop modal tests used to determine the receptances from excitation by an instrumented hammer. Experiment 2 (Assignment of zeros). To maintain geometrical symmetry of the T-plate identical zeros were assigned to h11 and h22 at m1,m2 107300i. Gains were determined as F 1.8I, G 3735I. Experimental open-loop and closed-loop receptances are shown in Fig. 10 where the frequency of the zeros is seen to agree very well with the assigned value of 300 rad/s and the poles remain stable. Experiment 3 (Simultaneous assignment of poles and zeros). Poles and zeros were assigned to h11 at l1,2 157295i and m1,2 127265i, respectively. Gain values of G diag(3720, 2355) and F diag (2.24, 5.75) were found and the resulting open-loop and closed-loop receptances were found as shown in Fig. 11. It can be seen in Fig. 11 that the frequencies of the rst peak and dip of the dashed (red) line, that represents the closed-loop receptance, agree closely with the imaginary part of the assigned poles and zeros. 5.1. Stability robustness In this section the stability robustness of the experimental T-plate structure to active vibration control by pole-zero placement is addressed. The problem considered is that of Example 2 in Section 4 above, namely the assignment of zeros of h11 and h22 to m1,m2 107300i. The open-loop transfer function between the input
30 20 10 0 10 20 30 40 50 60 0 100 200 300 400 Frequency (Hz) 500 600 700

Fig. 13. Spectrum of eigenvalues l1 (dashed line) and l2 (solid line) of the open-loop transfer function 0.37 H(io)B diag(gi+iofi)BT.

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1405

voltage to the actuators and the output displacement y(s) is then dened as 0.37 H(s)B diag(gi+sfi)BT, where the gain of 0.37 represents the actuator dynamic as described previously. Open-loop experiments were carried out by applying random voltages to the actuators in the range of 01000 Hz and measuring the output voltages, calibrated for displacement, from the dSPACE board. The frequency-domain requirements for the stability of single-inputsingle-output (SISO) systems take the form of the standard Nyquist criterion and in the MIMO case they involve its multivariable generalisation [30]. Thus the system may be considered stable if det[I+0.37 H(io)B diag(gi+iofi)BT], 0oooN, does not enclose the point (0, 0). It can be seen from Figs. 12(a) and (b) that this appears to be the case, although the system is extremely close to being marginally stable. The spectrum in Fig. 13 shows good roll-off of the eigenvalues l1 and l2 of the open-loop transfer function, indicated by the dashed line (red) and solid line (blue), respectively, thereby conrming that in practice spillover does not lead to instability at higher frequencies.
(a) 3.5 3 Minimum singular value 2.5 2 1.5 1 0.5 0 0 20 40 60 80 100 Frequency (Hz) 120 140 160

(b) 102

Receptance (Amplitude)

101

Openloop response Closedloop response (MSV > 0.6) Closedloop response (MSV > 0.7)

100 200 250 300 Frequency (rad/s) 350 400

Fig. 14. Robustness (a) minimum singular value. Solid linewithout constraint; dashed lineafter added constraint. (b) Receptance. Solid lineopen-loop receptance; dashed lineclosed-loop receptance for s 40:6 from 50 to 100 Hz; dot-dashed lineclosed-loop receptance for s 40:7 from 50 to 100 Hz.

ARTICLE IN PRESS
1406 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408

The robustness of the closed-loop system can be improved by increasing the minimum singular value s of [I+0.37 H(s)B diag(gi+sfi)BT] as explained by Maciejowski [31], thereby dening the following constraint: s I 0:37 HsB diaggi sf i BT 4r on the solution of the nonlinear characteristic equations. In our particular example, a constraint of s 40:7 was applied over the frequency range of 50100 Hz. The resulting control gains were found to be, ! ! 12:7 0 3734:9 0 F ; G . 0 13:8 0 3734:9

(a)

4 3.5 3 2.5 2 1.5 1 0.5 0 0 20 40 60 80 100 Frequency (Hz) 120 140 160

(b) 102

Receptance (Amplitude)

Minimum singular value

101

Openloop receptance Closedloop with constraint (MSV > 0.8) Closedloop MSV > 0.9 (50 100 Hz) and MSV > 0.45 (6 10 Hz)

100 200 250 300 350 400 Frequency (rad/s)

Fig. 15. Robustness (a) minimum singular value. Solid linewithout constraint; dashed lineafter added constraint. (b) Receptance. Solid lineopen-loop receptance; dashed lineclosed-loop receptance for s 40:8 from 50 to 100 Hz; dot-dashed lineclosed-loop receptance for s 40:9 from 50 to 100 Hz and s 40:45 from 6 to 10 Hz.

ARTICLE IN PRESS
J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 1407

In Fig. 14(a) the solid (blue) line denotes s versus frequency without constraint. The dashed (red) line is the case of s 40:7 for the range of 50100 Hz. A series of modal tests using hammer excitation were carried out with the feedback control gains obtained with the added constraints s 40:6 and 0:7. The solid (blue) line in Fig. 14(b) represents the unconstrained open-loop receptance. The dashed (red) and dot-dashed (green) lines represent the closed-loop receptances after the added constraints s 40:6 and 0:7 were applied over the range of 50100 Hz. It is seen that the effect of the constraints is to add damping and thereby improve the stability robustness of the T-plate system. Of course, this is achieved at the cost of reduced accuracy in the placement of the zeros. A further constraint, s 40:45, was then added to the range 610 Hz to address the sharp dip in the dashed line of Fig. 14(a) at low frequencies. In the range 50100 Hz the lowest singular value was constrained such that s 40:9. Control gains were then obtained as ! ! 17:35 0 1709:5 0 F ; G . 0 22:64 0 2164 Fig. 15 shows the minimum singular values and the closed-loop receptances for the system, now with two constraints. This results in considerably increased damping over the previous test and further deterioration in the accuracy of placing the zeros. 6. Conclusions Active vibration suppression by eigenvalue assignment using output feedback and the receptance method is presented. The method does not require the usual M, C, K matrices but uses measured receptances from the open-loop system instead. Illustrative numerical examples are presented, as are the results of a physical experiment using collocated inertial actuators and accelerometers, to assign poles and zeros (natural frequencies and anti-resonances). In the experiments, receptances H(s) are determined from the measured H(io) by tting rational fraction polynomials. Closed-loop receptances, obtained by applying gains determined from the analysis, have peaks and dips that agree very closely with the imaginary parts of the assigned poles and zeros. The stability robustness may be improved by applying a constraint to the singular values of the matrix return difference. Acknowledgements The authors wish to acknowledge helpful conversations with Professor P. Gardonio (Southampton University), Dr. A.T. Shenton (Liverpool University), Dr. R. Stanway (Shefeld University) and Professor G.-F. Yao (Jilin University). References
[1] W.M. Wonham, On pole assignment in multi-input controllable linear systems, IEEE Transactions on Automatic Control AC-12 (1967) 660665. [2] E.J. Davison, On pole assignment in linear systems with incomplete state feedback, IEEE Transactions on Automatic Control AC-15 (1970) 348351. [3] H. Kimura, Pole assignment by gain output feedback, IEEE Transactions on Automatic Control AC-20 (1975) 509516. [4] B.C. Moore, On the exibility offered by state feedback in multivariable systems beyond closed loop eigenvalue assignment, IEEE Transactions on Automatic Control AC-21 (1976) 689692. [5] J. Kautsky, N.K. Nichols, P. Van Dooren, Robust pole assignment in linear state feedback, International Journal of Control 41 (5) (1985) 11291155. [6] B.N. Datta, S. Elhay, Y.M. Ram, Orthogonality and partial pole placement for the symmetric denite quadratic pencil, Linear Algebra and its Applications 257 (1997) 2948. [7] I. Bucher, S. Braun, The structural modication inverse problem, Mechanical Systems and Signal Processing 7 (1993) 217238. [8] J.E. Mottershead, C. Mares, M.I. Friswell, An inverse method for the assignment of vibration nodes, Mechanical Systems and Signal Processing, Special Issue on Inverse Methods in Structural Dynamics 15 (1) (2001) 87100. [9] J.E. Mottershead, Structural modication for the assignment of zeros using measured receptances, Transactions of the American Society of Mechanical Engineers, Journal of Applied Mechanics 68 (5) (2001) 791798.

ARTICLE IN PRESS
1408 J.E. Mottershead et al. / Journal of Sound and Vibration 311 (2008) 13911408 [10] J.E. Mottershead, A. Kyprianou, H. Ouyang, Structural modication, part 1: rotational receptances, Journal of Sound and Vibration 284 (12) (2005) 249265. [11] A. Kyprianou, J.E. Mottershead, H. Ouyang, Structural modication, part 2: assignment of natural frequencies and antiresonances by an added beam, Journal of Sound and Vibration 284 (12) (2005) 267281. [12] J.E. Mottershead, M.G. Tehrani, D. Stancioiu, S. James, H. Shahverdi, Structural modication of a helicopter tailcone, Journal of Sound and Vibration 298 (12) (2006) 366384. [13] J.E. Mottershead, Y.M. Ram, Inverse eigenvalue problems in vibration absorption: passive modication and active control, Mechanical Systems and Signal Processing 20 (1) (2006) 544. [14] L. Benassi, S.J. Elliott, P. Gardonio, Active vibration isolation using an inertial actuator with local force feedback control, Journal of Sound and Vibration 276 (2004) 157179. [15] R.L. Clark, W.R. Saunders, G.P. Gibbs, Adaptive Structures, Dynamics and Control, Wiley/Interscience, New York, 1998. [16] P. Gardonio, E. Bianchi, S.J. Elliot, Smart panel with multiple decentralised units for the control of sound transmission. Part I: theoretical predictions, Journal of Sound and Vibration 274 (2004) 163192. [17] P. Gardonio, E. Bianchi, S.J. Elliot, Smart panel with multiple decentralised units for the control of sound transmission. Part II: design of decentralised control units, Journal of Sound and Vibration 274 (2004) 193213. [18] E. Bianchi, P. Gardonio, S.J. Elliot, Smart panel with multiple decentralised units for the control of sound transmission. Part III: control system implementation, Journal of Sound and Vibration 274 (2004) 215232. [19] R. Stanway, J.A. Rongong, N.D. Sims, Active constrained layer damping: a state-of-the-art review, Proceedings of the IMechE, Part I, Journal of Systems and Control 217 (6) (2003) 437456. [20] M.J. Balas, Feedback control of exible structures, IEEE Transactions on Automatic Control AC-23 (1978) 673679. [21] L. Meirovitch, Dynamics and Control of Structures, Wiley, New York, 1990. [22] K. Jian, M.I. Friswell, Designing distributed modal sensors for plate structures using nite element analysis, Mechanical Systems and Signal Processing 20 (8) (2006) 22902304. [23] F. Tisseur, K. Meerbergen, The quadratic eigenvalue problem, SIAM Review 43 (2) (2001) 235286. [24] Y.-H. Lim, S.V. Gopinathan, V.V. Varadan, V.K. Varadan, Finite element simulation of smart structures using an output feedback controller for vibration and noise control, Smart Materials and Structures 8 (1999) 324337. [25] U. Stobener, L. Gaul, Active vibration control of a car body based on experimentally evaluated modal parameters, Mechanical Systems and Signal Processing 15 (1) (2001) 173188. [26] Y.M. Ram, J.E. Mottershead, Receptance method in active vibration control, American Institute of Aeronautics and Astronautics Journal 45 (3) (2007) 562567. [27] A. Preumont, Vibration Control of Active Structures, second ed., Kluwer Academic Publishers, Dordrecht, 2002. [28] C.R. Fuller, S.J. Elliot, P.A. Nelson, Active Control of Vibration, Academic Press, San Diego, 1996. [29] D. Formenti, M. Richardson, Parameter estimation from frequency response measurements using rational fraction polynomials, Proceedings of the First International Modal Analysis Conference, Orlando, FL, 1982. [30] H.H. Rosenbrock, The stability of multivariable systems, IEEE Transactions on Automatic Control AC17 (1972) 105107. [31] J.M. Maciejowski, Multivariable Feedback Design, Addison-Wesley, Wokingham, England, 1989.